• Journal of Korea Multimedia Society
• /
• v.22 no.10
• /
• pp.1149-1159
• /
• 2019

We present a novel deep learning architecture for obtaining a latent image from a single blurry image, which contains dynamic motion blurs through object/camera movements. The proposed architecture consists of two sub-modules: blur image restoration and optical flow estimation. The tasks are highly related in that object/camera movements make cause blurry artifacts, whereas they are estimated through optical flow. The ablation study demonstrates that training multi-task architecture simultaneously improves both tasks compared to handling them separately. Objective and subjective evaluations show that our method outperforms the state-of-the-arts deep learning based techniques.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.497-500
• /
• 2022

High-quality images and videos are being generated as technologies for deep learning-based image style translation and conversion of static images into dynamic images have developed. However, it takes a lot of time and resources to manually transform images, as well as professional knowledge due to the difficulty of natural image transformation. Therefore, in this paper, we study natural style mixing through a style conversion network using GAN and natural dynamic image generation using the First Order Motion Model network (FOMM).

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.59 no.3
• /
• pp.651-656
• /
• 2010

This paper proposes a new dynamic PCA algorithm to recognize types of electric poles, which is necessary for a mobile robot moving along the neutral line for inspecting high-voltage facilities. Since the mobile robot needs to pass over the electric poles and grasp the neutral wire again for the next region inspection, the detection of the electric pole type is a critical factor for the successful passing-over the electric pole. The CCD camera installed on the mobile robot captures the image of the electric pole while it is approaching to the electric pole. Applying the dynamic PCA algorithm to the CCD image, the electric pole type has been classified to provide the stable grasping operation for the mobile robot. The new dynamic PCA algorithm replaces the reference image in real time to improve the robustness of the PCA algorithm, adjusts the brightness to get the clear images, and applies the Laplacian edge detection algorithm to increase the recognition rate of electric pole type. Through the real experiments, the effectiveness of this proposed dynamic PCA algorithm method using Laplacian edge detecting method has been demonstrated, which improves the recognition rate about 20% comparing to the conventional PCA algorithm.

• 한국정보컨버전스학회:학술대회논문집
• /
• 2008.06a
• /
• pp.145-150
• /
• 2008

This thesis explores the problem of reconstructing a three-dimensional(3D) scene given a set of images or image sequences of the scene. It describes efficient methods for the 3D reconstruction of static and dynamic scenes from stereo images, stereo image sequences, and images captured from multiple viewpoints. Novel methods for image-based and volumetric modelling approaches to 3D reconstruction are presented, with an emphasis on the development of efficient algorithm which produce high quality and accurate reconstructions. For image-based 3D reconstruction a novel energy minimisation scheme, Iterated Dynamic Programming, is presented for the efficient computation of strong local minima of discontinuity preserving energyy functions. Coupled with a novel morphological decomposition method and subregioning schemes for the efficient computation of a narrowband matching cost volume. the minimisation framework is applied to solve problems in stereo matching, stereo-temporal reconstruction, motion estimation, 2D image registration and 3D image registration. This thesis establishes Iterated Dynamic Programming as an efficient and effective energy minimisation scheme suitable for computer vision problems which involve finding correspondences across images. For 3D reconstruction from multiple view images with arbitrary camera placement, a novel volumetric modelling technique, Embedded Voxel Colouring, is presented that efficiently embeds all reconstructions of a 3D scene into a single output in a single scan of the volumetric space under exact visibility. An adaptive thresholding framework is also introduced for the computation of the optimal set of thresholds to obtain high quality 3D reconstructions. This thesis establishes the Embedded Voxel Colouring framework as a fast, efficient and effective method for 3D reconstruction from multiple view images.

• Proceedings of the IEEK Conference
• /
• 2008.06a
• /
• pp.913-914
• /
• 2008

This paper presents a method of adaptive image enhancement with dynamic range compression and contrast enhancement. The dynamic range compression is to adaptively enhance the dark area using illumination component of DCT compression block. The contrast enhancement is to modify the image contrast using retinex theory that uses the HVS properties. The block artifacts and other noises, caused by processing in the compression domain, were removed by after processing.

• Journal of the Earthquake Engineering Society of Korea
• /
• v.13 no.3
• /
• pp.11-19
• /
• 2009

Recently, concerns relating to the maintenance of large structures have been increased. In addition, the number of large structures that need to be evaluated for their structural safety due to natural disasters and structural deterioration has been rapidly increasing. It is common for the structural characteristics of an older large structure to differ from the characteristics in the initial design stage, and changes in dynamic characteristics may result from a reduction in stiffness due to cracks on the materials. The process of deterioration of such structures enables the detection of damaged locations, as well as a quantitative evaluation. One of the typical measuring instruments used for the monitoring of bridges and buildings is the dynamic measurement system. Conventional dynamic measurement systems require considerable cabling to facilitate a direct connection between sensor and DAQ logger. For this reason, a method of measuring structural responses from a remote distance without the mounted sensors is needed. In terms of non-contact methods that are applicable to dynamic response measurement, the methods using the doppler effect of a laser or a GPS are commonly used. However, such methods could not be generally applied to bridge structures because of their costs and inaccuracies. Alternatively, a method using a visual image can be economical as well as feasible for measuring vibration signals of inaccessible bridge structures and extracting their dynamic characteristics. Many studies have been conducted using camera visual signals instead of conventional mounted sensors. However, these studies have been focused on measuring displacement response by an image processing technique after recording a position of the target mounted on the structure, in which the number of measurement targets may be limited. Therefore, in this study, a model experiment was carried out to verify the measurement algorithm for measuring multi-point displacement responses by using a DIC (Digital Image Correlation) technique.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.52 no.1
• /
• pp.111-120
• /
• 2015

HDR(High dynamic range) imaging is a technique to represent a dynamic range of real world. Exposure fusion is a method to obtain a pseudo-HDR image and it directly fuses multi-exposure images instead of generating the true-HDR image. However, it results ghost artifacts while fusing the multi-exposure images with moving objects. To solve this drawback, temporal consistency assessment is proposed to remove moving objects. Firstly, multi-level threshold bitmap and brightness bitmap are proposed. In addition, hue-angle constancy map between multi-exposure images is proposed for compensating a bitmap. Then, two bitmaps are combined as a temporal weight map. Spatial domain image quality assessment is used to generate a spatial weight map. Finally, two weight maps are applied at each multi-exposure image and combined to get the pseudo-HDR image. In experiments, the proposed method reduces ghost artifacts more than previous methods. The quantitative ghost-free evaluation of the proposed method is also less than others.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38A no.8
• /
• pp.669-679
• /
• 2013

When an image is converted from HDR (high dynamic range) to LDR (low dynamic range), a tone mapping process is the essential component. Many TMOs (tone mapping operators) have been motivated by human vision which has lower physical luminance range than that in real scene. The representative of human vision properties which motivate TMOs is the local adaptation. However, TMOs are ultimately compressing image information such as contrast, saturation, etc. and the compression causes defects in image quality. In this paper, in order to compensate the degradation of the image which is caused by TMOs, the visual acuity-based edge stop function is proposed for applying the property of human vision to base-detail separation. In addition, using CSF (contrast sensitivity function) which represents the relationship among spatial frequency, contrast sensitivity, and luminance, the sharpness filter is designed and adaptively applied to the detail layer in regard to surround luminance.

• Korean Journal of Remote Sensing
• /
• v.26 no.4
• /
• pp.395-401
• /
• 2010

Image enhancement algorithm aims to improve the visual quality of low contrast image through eliminating the noise and blurring, increasing contrast, and raising detail. This paper proposes adaptive dynamic range linear stretching(ADRLS) algorithm based on advantages of existing methods. ADRLS method is focused on generating sub-histograms of the majority through partitioning the histogram of input image and applying adaptive scale factor. Generated sub-histograms are finally applied by linear stretching(LS) algorithm. In order to validate proposed method, it is compared with LS and histogram equalization(HE) algorithm generally used. As the result, the proposed method show to improve contrast of input image and to preserve distinct characteristics of histogram by controlling excessive change of brightness.

• Journal of IKEEE
• /
• v.22 no.4
• /
• pp.901-906
• /
• 2018

Due to the development of display technologies for images, the resolution and quality of images are increasing day by day. In accordance with the development of the display technology, researches have been actively conducted on technologies for converting and displaying existing images to higher resolution and quality. Since the results of theses studies are included in the image signal processor, hardware implementation is indispensable. In this paper, we propose a real-time HDR(High Dynamic Range) conversion hardware implementation of LDR(Low Dynamic Range) image using luminance distribution. The proposed method extracts the features of the image using the histogram of the luminance distribution, and extends the luminance and color based on the extracted features. In addition, when the proposed method is designed by hardware IP(Intellectual Property) and its performance is verified, 4K DCI(Digital Cinema Image) can be handled at a rate of 30fps at 265.46MHz.